System and Method for Identifying Items

1. A method for item identification, comprising: determining an image set of a set of items, wherein the image set is captured by a set of cameras; generating a point cloud for the set of items; generating a height map of the items using the point cloud; determining a region mask for each item based on the height map using a segmentation classifier; generating a coarse mesh for each item using the respective region mask; determining an image segment set for each item by projecting the respective coarse mesh into each image of the image set; and determining an item identifier for each item using the respective image segment set.

2. The method of claim 1 , wherein the item identifier is determined for each image within 1 second of image capture.

3. The method of claim 1 , wherein the set of cameras comprise stereo camera pairs and a structured light camera.

4. The method of claim 1 , further comprising determining calibration matrices, wherein the calibration matrices are used to generate the point cloud from the image set and to reproject each coarse mesh into the respective image.

5. The method of claim 1 , wherein the point cloud is quantized.

6. The method of claim 1 , wherein the coarse mesh is calculated using a convex hull algorithm.

7. The method of claim 1 , wherein determining the image segment set for each item comprises mapping cell coordinates for each cell of the respective coarse mesh to color image pixel locations of each color image from the image set to determine pixels of each image that correspond to each item.

8. The method of claim 7 , wherein determining the image segment set further comprises assigning pixels of an image to image segments for items associated with coarse mesh cells closest to a camera frame of the respective camera.

9. The method of claim 1 , wherein the height map comprises a maximum height within the point cloud along a z-axis for each x-y position.

10. The method of claim 1 , wherein the segmentation classifier can be an instance-based segmentation classifier.

11. The method of claim 1 , wherein the point cloud is generated using the image set.

12. The method of claim 1 , wherein the coarse mesh is generated based on the height map.

13. The method of claim 12 , wherein the coarse mesh is generated by projecting a height map segment for an item, defined by a region mask for the item, downward along the z-axis.

14. The method of claim 1 , wherein the image set comprises images captured from at least three of a front, back, left, or right view of the measurement volume.

15. The method of claim 1 , wherein determining the item identifier comprises determining an item identifier for each image segment set, using a neural network.

16. The method of claim 15 , wherein determining the item identifier comprises: determining candidate item identifiers for each image segment using a convolutional neural network; and determining the item identifier based on the candidate item identifiers using a majority voting algorithm.

17. A system for item identification, comprising: a set of sensors configured to sample measurements of a set of items within a measurement volume, wherein the set of sensors comprise a set of cameras, each camera is associated with a camera field of view, and the measurements comprise a set of images captured by the set of cameras; and a processing system configured to: generate a point cloud using the measurements; generate a height map using the point cloud; determine a region mask for each item using a segmentation classifier that ingests the height map as input; generate a coarse mesh for each item using the respective region mask; determine an image segment set for each item by projecting the respective coarse mesh into each image of the set of images; and determine an item identifier for each item using the respective image segment set.

18. The method of claim 17 , wherein the set of cameras are statically mounted to a common housing, wherein the common housing comprises a static base defining a bottom of the measurement volume.

19. The method of claim 18 , wherein the base comprises a calibration pattern, wherein each camera field of view encompasses the calibration pattern, wherein the processor is further configured to periodically determine calibration matrices for the cameras based on the calibration pattern, wherein the calibration matrices are used to project the coarse mesh of each item into each image of the set of images.

20. The method of claim 18 , wherein the processing system is mounted to the common housing.